Use CHECK instead of PCHECK, because PCHECK is used for errno.

The implementation of BarrierStat is buggy, and it is not necessary
for Paddle to diagnose which node in cluster is slow.

Fix #2567

* Also fix a protential bug in WarpCTCLayer.cpp

Also fix compile issues

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Because in cluster maybe use a lot machine to train a model, and some parameter size could be too small for ParameterServer. Then some of pservers could not have any ParamBlock.
* Also, because ports_num or ports_num_for_sparse is too large, then give a warning in runtime.

* fix DataProvider create function args bug

Change-Id: I9e3a1c535c805bf30204a14aea8d5143ff534784

* remove PserverForPython.h which is not used

Change-Id: I2b27f1f3c11a42766a92fc689f0f5f1f73ee1d70

* add internal document script

Change-Id: Ia0fec79456caea0b271f9903cc13e8a3d32e0774